Screening for asymptomatic disease as well as prediction and decision rules are two increasingly important applications of the principle of testing that we examined in Chapters 8 and 9. In this chapter, we will take a look at each of these applications of testing principles.
Criteria for Successful Screening
Screening is a special form of testing that aims to detect specific diseases in individuals who are asymptomatic for that disease. The goal of screening for a disease is to identify asymptomatic individuals who have the disease in order to intervene to improve outcome.10.1
When evaluating a screening test, it is helpful to utilize a series of ideal criteria for justifying screening. Although few screening tests fulfill all of the following criteria, these criteria provide an ideal standard against which particular screening tests can be compared:
- Substantial morbidity and mortality: The disease or condition often leads to death or disability.
- Early detection improves outcome: Early detection is possible and improves outcome.
- Screening is feasible: A high-risk group can be identified and tested using a testing strategy with good diagnostic performance.
- Screening is acceptable and efficient: The testing strategy has acceptable harms, costs, and patient acceptance.
Let us see how we can use these criteria to evaluate the use of screening tests.
Substantial Morbidity and Mortality
The importance of selecting diseases for screening that produce substantial morbidity and mortality is the key starting point for screening. Morbidity may include disabilities such as blindness or strokes, or extended period of costly health care such as kidney dialysis or treatment for coronary artery disease. Despite the importance of identifying conditions for screening that produce substantial morbidity and/or mortality, this condition may be ignored, as illustrated in the following example:
Screening for asymptomatic hemorrhoids is being considered all adults. The condition is found to have a high prevalence. However, screening was not recommended since asymptomatic hemorrhoids was also found to result in little morbidity or mortality.
Despite the high prevalence of asymptomatic hemorrhoids, the low morbidity and mortality makes it a poor candidate for screening. A common condition that poses little harm to individuals is not a good candidate for screening.
10.1 Asymptomatic implies that the individual does not have symptoms of the disease for which the screening test is being used. They may have other diseases and/or other symptoms. The term screening may be used with other somewhat different meanings. Tests may be used in the presence of symptoms when the clinician wishes to test for various physiological measurements or a range of possible diseases. Screening may also refer to a panel of tests designed to differentiate the cause of a clinical pattern, such as drug screening in the presence of clinical manifestations of intoxication. Screening for asymptomatic disease should also be distinguished from case finding. Case finding usually, but not always, refers to identification of an individual with a communicable disease with the intention of locating and treating their contacts.
Early Detection Improves Outcome
The evidence that supports the ability to detect disease at an early stage often comes from studies that compare the stage of disease among individuals diagnosed through screening with those whose disease was diagnosed in the usual course of health care. The probabilities of detecting disease in early stages through screening and through the usual course of health care may be compared. If there is a higher probability of detecting early disease with screening, the results suggest early detection is possible through screening of asymptomatic individuals, as illustrated in the next example:
An investigation of the diagnosis of breast cancer through monthly self-examination and yearly clinical examination was compared with the diagnosis of breast cancer using these methods as well as yearly mammography among women 50 years of age and older. The investigation demonstrated that the use of mammography increased the probability that the diagnosis of breast cancer would occur at an earlier state of the disease.
Achieving early detection, however, is not necessarily the same as detecting disease that will go on to cause morbidity or mortality. It is possible that the disease detected by screening may never become clinically important, as illustrated in the next example:
A new test is able to detect thyroid cancer in 20% of all men older than 80 years. Cancers detected in these men using the new test are generally found to be microscopic foci that are at an earlier stage than thyroid cancers diagnosed during the course of health care. The investigators are enthusiastic about the possibility of early detection of thyroid cancer and argue that this test is likely to be useful in screening for thyroid cancer in men older than 80 years.
The ability to detect cancer early is not the same as the ability to detect cancers that are likely to go on to become clinically important. Older men may die with thyroid cancer rather than die from thyroid cancer. The goal of early detection is not just to identify cancer early, but also to identify those cases that need effective therapy to prevent progression to clinically important disease.
Screening should not be recommended unless an intervention is available that can improve the outcome of patients detected by screening. Thus, unless there is therapy, or other effective interventions, that is more effective when used early in the disease, there is generally no reason to conduct screening for disease. Thus, the ability to detect disease at an early stage is not enough to fulfill this second criterion for screening. Effective treatment must be available and be more effective when used during the asymptomatic phase.10.2
The benefit of screening is ideally demonstrated using a randomized controlled trial that randomizes patients to a screening group and a usual medical care control group.10.3
Often, however, it is not possible to perform randomized controlled trials with long-term follow-up. Thus, we often rely on studies that compare the outcome of groups that have been screened with that of groups that have not been screened by conducting cohort studies. These studies may provide important data that suggest the ability of screening to successfully improve the outcome.
Cohort studies of screening, however, are susceptible to misleading results because of lead-time bias. This bias results from comparing the time from diagnosis to an outcome, such as death, between those diagnosed through screening and those diagnosed in the usual course of health care. The potential for lead-time bias is illustrated in the next example:
An X-ray screening program to detect lung cancer was performed among a group of smokers who were asked to participate. Their outcomes were compared with the outcomes of individuals in a control group whose lung cancer was diagnosed in the usual course of medical care. The study and control groups’ individuals were matched for age and number of pack-years of cigarette smoking. The screened group had a greatly improved survival 1 year after their diagnosis of lung cancer compared with the survival 1 year after diagnosis among the unscreened control group.10.4
Even if the treatment for lung cancer has no effect, we would expect the results for the screened group to be better. By detecting the disease earlier, screening with chest X-rays moves back the time of diagnosis. As illustrated in Figure 10.1, unfortunately, screening with chest X-rays has not moved forward the time of death. The increase in time between diagnosis and death may be entirely due to lead-time bias; that is, the screening had led to early detection without improved prognosis.
FIGURE 10.1. Lead-time bias in which earlier diagnosis by screening does not alter outcome.
There is a second reason why comparing screened and unscreened populations using a cohort study to assess their outcome may not produce convincing evidence of an improved outcome among those screened. This is known as length bias. As illustrated in Figure 10.2, length bias occurs when there are two different types of a disease, one of which is slowly progressing and one of which is rapidly progressive such as slow-growing and rapidly growing cancer.
FIGURE 10.2.Length bias demonstrating why more slowly progressive cases of disease may be detected by screening.
Solid lines indicate preclinical phase; dotted lines, clinical phase; and circles, death or other end point.
When screening is performed initially, most cases that are detected will be slow growers. This is because slow growers remain in the presymptomatic or asymptomatic stage for a longer period of time and thus constitute most of cancer cases detected by screening. Fast growers, on the other hand, remain in the presymptomatic stage for a shorter period of time and constitute a smaller proportion of cases of cancer detected by screening. When length bias results in detecting mostly slow growers with a good prognosis, it appears that screening has done its job of improving outcome. Unfortunately, all that has happened is the identification of a group with good prognosis who have a better outcome than a group with poor prognosis made up mostly of rapid growers.10.5
Length bias, like lead-time bias, can be prevented by the use of large randomized controlled trials since we can assume that the study and control groups have the same proportion of slow growers and rapid growers. Let us take a look at an example that illustrates the advantage of a randomized controlled trial:
A study of prostate cancer among men older than 70 years compared the time between diagnosis and death for men screened for prostate cancer with those who were diagnosed in the course of medical care. Prostate cancers have a wide range of prognosis depending on the rate of growth of the cancer. The investigator found that screening produced impressive increases in the time between diagnosis and death compared with diagnosis among those who are not screened. When a randomized controlled trial was conducted, the increased longevity for the screened group and the unscreened group was nearly identical.
This example illustrates the potential for length bias whenever we are dealing with disease with a range of rates of progression. Conducting randomized controlled trials of screening may at times be the only effective way of taking length bias into account.
For diseases or conditions that cause substantial morbidity or mortality and where early detection improves outcome, we would ideally like to be able to provide screening to detect asymptomatic disease. However, before this can be advocated, two additional criteria should be fulfilled: Screening is feasible and screening is acceptable and efficient.
10.2 At times, screening may be worthwhile for other reasons. It may be worthwhile to detect communicable disease in order to prevent spread even if no effective treatment is available
10.3 Even when using a randomized controlled trial, it is necessary to follow up those diagnosed with the disease. They should be monitored not just until they are diagnosed, but until they have had an opportunity to develop the adverse outcome we hope to prevent. That is because a randomized controlled trial that demonstrates improvement in early outcome is not always sufficient. The outcome in the screened group should remain better than groups undergoing the usual methods of diagnosing the disease, even years after the disease is detected.
10.4 In recent years, spiral computed tomography (CT) scanning has shown promising results for early detection and treatment. It may now be possible to find lung cancers at a stage where cure is still possible. The fact that spiral CT is able to detect far smaller lung cancer may mean that it can detect lung cancer at a stage in which early detection is not only possible, but can improve outcome. Despite the success of spiral CT, it needs to be recognized that a large number of false positives will occur
10.5 Length bias assumes that disease that slowly progresses in the presymptomatic stage will remain slowly progressive once it enters the symptomatic phase. Length bias can be partially taken into account by studying groups that have previously undergone screening, thus removing from the group most of the long-standing cases of the disease.
Screening Is Feasible
Need for a High-Risk Group and More Than One Test
As we have seen, Bayes’ theorem tells us that the pretest probability of a disease usually has a very strong relationship to the probability of disease after the results of the test are obtained. Thus, we need a screening strategy that allows us to identify a group at high enough risk of the disease and a testing approach that has a good enough diagnostic performance.
When performing screening, we are usually testing presymptomatic or asymptomatic individuals. Thus, we cannot rely on their symptoms to help us estimate the pretest probability of disease. Instead, we need to rely on the prevalence of the disease itself and the presence of risk factors to help us identify groups with adequately high pretest probabilities of disease.
Without being able to identify individuals who have one or more risk factors for the disease, we would often be starting with a very low pretest probability. In Chapter 9, we illustrated the posttest probabilities or predictive values when using a test with 80% sensitivity and 90% specificity on a population with 1%, 10%, 50%, and 90% probability of the disease before conducting the test. The 1% example was used to illustrate the pretest probability when risk factors for a common disease are present in a population to be screened. In this situation, it was evident that one test alone would not be adequate for diagnosis. Learn More 10.1 illustrates what may happen when the prevalence of the disease is even lower such as 1 per 1,000, which might represent the prevalence of the human immune deficiency virus (HIV) in a low risk or general population.
Learn More 10.1: Testing for HIV
When the pretest probability is considerably lower than 1% for instance in the range of 1 per 1,000, successful screening is very difficult to achieve. This is the situation even when a test with high sensitivity and high specificity is used, as illustrated in the next example: